Research And Application On Fault Diagnosis Of Non-Gaussian Stochastic System

With the rapid development of industrial technology, the requirements of industrial operational safety and reliability is increasing, fault diagnosis has attracted a lot of attention from scholars. Some approaches to fault diagnosis of deterministic system have been developed. However, in actual industrial processes, stochastic noise and disturbance are not necessarily obey Gaussian distribution, which increases the difficulty of system fault diagnosis. Thus, stochastic distribution theory is used to deal with fault diagnosis of non-Gaussian stochastic system.Firstly, considering non-Gaussian system noises and measurement noises of practical systems, a filter is designed based on maximum information potential criteria. To estimate the probability density function of the estimate error, generalized density evolution equations and Parzen window method are proposed. Moreover, the mean square bounded convergence analysis is given. Secondly, for wind energy conversion systems inevitably affected by random measurement noises and random wind speed, sensor fault diagnosis method of wind energy conversion system is proposed based on a bank of maximum information potential filters. Residual is produced by the proposed filter, thereby effectively determining sensor malfunction. Then a decision function is designed to locate faults. Simulation results show that this method can effectively diagnose sensor failure of wind energy conversion systems. In addition, a new fault diagnosis method is proposed to diagnose sensor fault and fouling of heat exchanger. This method fully consider the accuracy of the sensor data. It not only can detect failure, but also dignose fault. It can be seen that proposed method can effectively diagnose fouling fault and sensor fault of heat exchanger.